Comparison of GA-Shift Neighbourhood Mutation and GA-Pairs Exchange Mutation with Multi Cut Point Crossover in Solving RICH-VRP

  65

Comparison of GA-Shift Neighbourhood Mutation and GA-Pairs Exchange

Mutation with Multi Cut Point Crossover in Solving RICH-VRP

_{1) 2)}

  

Ismail Yusuf Panessai* , Sanusi ,

_{3) 4) 1)}

Pratiwi Hendro Wahyudiono , Nurdiana Daeng Pawawo

₂₎ Sekolah Tinggi Teknik Ibnu Sina. Jl. Teuku Umar, Lubuk Baja, Batam.

  Lamintang Education & Training (LET) Centre, Buana Impian Blok B1 No. 27 & 28. Batam E-mail:

  

Abstract

This research was focused on a heterogeneous fleet of passenger ships to solve multi

depot by using genetic algorithm (GA) to solve combinatorial problem i.e. vehicle routing

problem (VRP). The objective of this study is to compare the roulette wheel selection, multi cut

point crossover, and shift neighbourhood mutation with roulette wheel selection, multi cut point

crossover, and pairs exchange mutation to minimize the sum of the fuel consumption travelled,

the cost for violations of the ship draft and sea depth, and penalty cost for violations of the load

factor; maximize number port of call; and maximize load factor. Problem solving in this study is

how to generate feasible route combinations for rich VRP that meets all the requirements with

optimum solution. Route generated by roulette wheel selection, multi cut point crossover, and

shift neighbourhood mutation could decrease fuel consumption about 19.4350% compared to

roulette wheel selection, multi cut point crossover, and pairs exchange mutation about 18.6738%.

  Keywords — Vehicle Routing Problem; Genetic Algorithms; Multi Cut Point; Shift Neighbourhood Mutation; Pairs Exchange Mutation 1.

  INTRODUCTION Vehicle routing problem (VRP) is a classical combinatorial optimization problem. It is a key component of transportation management. It was first introduced to determine vehicle routes with minimum cost to serve a set of customers whose geographical coordinates and demands are known in advance [1]. A vehicle is required to visit each customer only once. Typically, vehicles are homogeneous and have the same capacity restriction.

  VRP can be represented as the following graph-theoretic problem. Let G = (P, A) be a complete graph where P = {

  0, 1, …, n} is the vertex set and representing customers with the depot

  located at vertex 0; A is the arc set. Vertices j = { _{j ij} 1, 2, …, n} correspond to the customers, each with a known non-negative demand, d . A non-negative cost, c , is associated with each arc (i, j) _{ij ji} ϵ A and represents the cost of travelling from vertex i to vertex j. If the cost values satisfy c

  ≠ c

  for all i, j ϵ P, then the problem is said to be an asymmetric VRP; otherwise, it is called a _ij symmetric VRP. In some contexts, c can be interpreted as a travel time or travel cost.

  The VRP consists of designing a set vehicle routes with minimum cost, defined as the sum of the costs of the routes’ arcs such that:  all vehicle routes start and end at the same depot  each customers in P is visited exactly once by exactly one vehicle  some side soft and hard constraints are satisfied

  MDVRP has been propose as follow each depot stores and supplies various products, and has a number of identical vehicles with the same capacity to serve customers who demand different quantities of various products [2]. Each vehicle starts the tour from its resided depot, delivers products to a number of customers, and returns to the same depot. One variant of the CVRP is the heterogeneous fleet vehicle routing problem (HVRP). In HVRP, the fleet is composed of a fixed number of vehicles with different in their equipment, capacity, age or cost

  Comparison of GA-Shift Neighbourhood Mutation and GA-Pairs Exchange 66 Mutation with Multi Cut Point Crossover in Solving Rich-VRP

  which the number of available vehicles is fixed as a priori [3]. The decision is how to best utilize the existing fleet to serve customer demands.

  Forward six formulations used by Yaman [4] which are enhanced by valid inequalities and lifting; Choi & Tcha [5] present a linear programming relaxation of which is solved by the column generation technique and used column generation technique which is enhanced by dynamic programming schemes; a branch-cut-and-price algorithm over an extended formulation that capable for solving HVRP proposed by Pessoa, et.al. [6] and a tabu search used approach using GENIUS for HVRP [7].

  Developing an algorithm based on heuristics and followed by a local search procedure based on the steepest descent local search and tabu search [8] while three phase heuristic developed by Dondo, et.al. [9] and an iterated local search base on heuristic proposed by Penna, et.al. [10]. Hybrid algorithm that composed by an iterated local search based on heuristic and a set partitioning formulation proposed by Subramanian, et.al. [11]. The set partitioning model was solved by means of a mixed integer programming solver that interactively calls the iterated local search heuristic during its execution.

  An evolutionary hybrid meta-heuristic which combines a parallel genetic algorithm with scatter search presented by Ochi, et.al. [12], while a record-to-record travel metaheuristic published by Li, et.al. [13]. In addition, memetic algorithm to solve HVRP is used by Prins [14].

  HVRP has been solved by implementing a threshold accepting procedure where a worse solution is only accepted if it is within a given threshold [15]; and provided an improved version in Tarantilis, et.al. [16]. A memory programming metaheuristic proposed by Li, et.al. [17] While tabu search algorithm to solve HFVRP used by Brandão [18].

  Several simple heuristic has been developed by Nag, et.al. [19] and more advanced heuristic proposed by Chao, et.al. [20], tabu search by Cordeau & Laporte [21] while memetic algorithm to solve SDCVRP used by Nagata & Bräysy [22].

  AVRP is related to Asymmetric Travelling Salesman Problem (ATSP). It is a generalized travelling salesman problem in which distances between a pair of cities need not equal in the opposite direction. The ATSP is an NP-hard problem, thus many meta-heuristic algorithms have been proposed to solve this problem, such as hybrid genetic algorithm by Choi, et.al. [23] and tabu search proposed by Basu, et.al. [24].

  2.VEHICLE ROUTING PROBLEM MODEL This study is on a heterogeneous fleet of passenger ships to solve multi depot. The objective of our problem is to consist of: i.

  Minimum fuel consumption The fuel consumption of each vehicle depends on the type is related with the type of engine used proposed by Ismail et. al. [25]:

  k k k k

  (1)

  f   P  

T

_k * * * * f _k = Total fuel consumption served by ship k

  T = Total voyage time by ship k k

  L = Total distance travelled for route r served by ship k _k r v = Speed of ship k η = High Speed Diesel constant (0.16) _k P = Engine power of ship k (HP)

  Φ = Number of engine

  Μ = Efficiency (0.8) ii.

  Maximum number port of call Number port of call of the route r that served by ship k donated by

  k  r Panessai, Sanusi, Wahyudiono, Pawawo 67 iii.

  Maximum load factor Load factor of ship k in each path calculated by: _k

  (2) _k



_ij

  b  _ij k

q

_k b = Load factor of ship k sailing from port i to port j in route r ij k

   = Number of passenger in ship k sailing from port i to port j in route r ij q k

  = Capacity of ship k Hard constraints are dealt with by removing unfeasible route. Hard constraints in this study include: i.

  Fuel Port A route must include at least one fuel port. ii.

  Travel Time The maximum duration for each tour is called commission days, T which is 14 days for this

  k k T T T T .

  case. Hence, ship must return to the depot within . If is the ship’s voyage time, then  iii.

  Travel Distance Since each ship has a different fuel tank size. Hence, total distance travelled for route r served

  k k L L

  by ship k, that it can travel is different. If is maximum allowed routing distance for ship

  r k k k L L L k , then  . calculated by: r _{k k k k} * v

     ( v _{k k} * 24 ) (3) L P  * * *

   

  Where, _k

  L _k = Maximum allowed routing distance for ship k = Maximum capacity tank of the ship k θ _k v = Speed of ship k

  η = High Speed Diesel constant (0.16) _k P _k = Engine power of ship k (HP)

  Φ = Number of engine used in ship k

  μ = Efficiency (0.8)

  A. Mathematical model

  Let, G = (P, A) be a graph, where P is the set of all ports, denoted by the nodes C (customer ports) and D (fuel ports) at which K is a set mix vehicles with capacity q k are based.

  A = {(i, j) │

i, j; i < j} is the set of arcs. Every arc (i, j) is associated with a non-negative distance matrix L=

_k l , ^ij _ij which represents the asymmetric travel distance from port i to port j, i.e., l may be different _ji from l ; i, j P.

  ∈

  In order to present the mathematical formulation of the models, we use the following:

  B. Notation C

  = {1, 2 , …, m} is a set of customer ports

  D = {(m+1), (m+2)

  , …, (m+n)} is a set of fuel ports

  

P = C  = {1, 2 D , …, m, (m+1), (m+2), …, (m+n)} is the set of all ports; n(P) = number of the

  ports

  K = {1, 2 , …, k} is a set of ship; n(K) = number of the ships

  C. Parameter _i h = Sea depth of port i, i

  ∈ {1, 2, …, m+n } Comparison of GA-Shift Neighbourhood Mutation and GA-Pairs Exchange 68 Mutation with Multi Cut Point Crossover in Solving Rich-VRP k v = Speed of ship i k

   = Ship draft of ship i; i

  ∈ {1, 2… n (K)}

  k r

  = Route i for ship k

  i k f

  = Fuel consumption for ship k to sail from port i to port j

  ij k f = Fuel consumption for ship k to serve route r r k t = Travel time for ship k sailing from port i to port j ij k

  T = Travel time for ship k sailing from port i to port j and stay in port i ij k

  T = Total voyage time by ship k

  T = Maximum allowed routing time (commission days) k l

  

= Distance travelled for ship k sailing from port i to port j; l ij may be different from l ji

ij k

  L =

  Distance travelled for ship k sailing from port i to port j and back to port i

  ij k L =

  Total distance travelled for route r served by ship k

  r k L =

  Maximum allowed routing distance for ship k

  k b

  = Load factor for ship k sailing from port i to port j

  ij k B

  = Average load factor for ship k sailing from port i to port j and back to port i

  ij k B = Average load factor for route r served by ship k _k r q = Available seat capacity of the ship k travel from ports i to j _{ij k}

   = Number of passenger on board, travel from ports i to j _ij

  α = Penalty cost for violations of the ship draft and sea depth β = Penalty cost for violations of the load factor

  ξ = Number port of call

D. Decision variables

  k i _{k } 1 if ship is used for serving route u  _i  otherwise 

  1 if port i is served by ship k in route r _k  w  _{r i }

  , otherwise 

  500 if ship k with ship draft sailing from port i with sea depth h , where  h  δ _{k i k i}     otherwise  _k

  B

β is a variable for ship k sailing from port i to port j and back to port i by average load factor ij

_k  B  1000 _ij

  50  _k   B 

   500

  50

  75  ij  otherwise 

  The problem is to construct route with minimum fuel consumption in feasible set of routes for each vehicle. The feasible route for ship k is to serve ports without exceeding the constraints:

  k T T

  1. Total travel time for any vehicle is no longer than

  k k L L

  2. Total travel distance for any vehicle is no longer than

  i Panessai, Sanusi, Wahyudiono, Pawawo 69

  3. The feasible route must include at least one fuel port The mathematical formulation is given in: _{k k k k}

    minimize f . u  . u  . u (4) _{ij ij ij ij k  K i  P j  P k  K i  P j  P k  K i  P j  P}          _k

  (5)

  maximize  _r

   _k maximize B (6) _r

  

  1. All ports (customer and fuel port) i are serviced by ship k minimum at once _k (7)

  u   i  P  k  K _ij 1 , ,   _{k  K i  P} 2. Travel time of the ship k is no longer than the maximum allowed routing time T , T = 14 days. _k T T

    _{k  K (8)}

  3. Total distance travelled for route i served by ship k is no longer than the maximum allowed

  k k L L .

  routing distance of the ship k, then 

  i _{k k}

  (9) 

  L L _{i k  K}  _k v

  4. Travel time of ship k equals to the distance travelled and divided by running speed . _k

_k

L (10)

  T  _k v

  5. The vehicle capacity constraint _{k k} (11)

  q . u  q _{i ij ij}    _{k K i P j P k i}   

  6. Ship k with ship draft sailing from port i with sea depth h and it is equal to

  δ _k α.

  

  (12)

   x  _{hi k  K i  P j  P}    k

  B

  7. Ship k sailing from port i to port j and back to port i by average load factor and it is equal

  ij

  to β. _k (13)

   B  _{ij k  K i  P j  P}   

  8. Route r served by ships k should possess a fuel-port _k (14)

  p u  . _ij

  1 _{k  K i , j  P p  D}   

  Three objectives in our study are minimum fuel consumption, maximum port of call and maximum load factor. All objective tested in different scenarios.  Minimum fuel consumption

  In this case the fitness value is the total fuel consumption of each ship. Obviously it is a minimization problem, thus the smallest value is the best. The fitness function represents as Eq. (15):

  1

• f 

  _k

  1,000,000

  (15)

  1  k f

  

f 

_r

  = Fuel consumption for ship k to serve route r

  r

  Comparison of GA-Shift Neighbourhood Mutation and GA-Pairs Exchange 70 Mutation with Multi Cut Point Crossover in Solving Rich-VRP

   Maximum number port of call In this case the fitness value is the total port of call of each route. Obviously it is a maximization problem, thus the largest value is the best. The fitness function represents as Eq. (16): _k

  (16)

  

f  

_r  k

   = Number port of call of the route r that served by ship k r

   Maximum load factor In this case the fitness value is the average of load factor of each route. Obviously it is a maximization problem, thus the largest value is the best. The fitness function represents as Eq. (17): _k (17)

  

f  B

_r



k

  B

  = Average load factor for route r served by ship k

  r

  In this research, a classical selection method is used for solving routing problems, namely the roulette wheel selection. Selection process begins by spinning the roulette wheel n times; each time, a single chromosome is selected for a new population in the following 2 steps: Step 1: Generate a random number r in a range [0, 1]. _{1 1} Step 2: If , then select the first chromosome s otherwise, select the s-th chromosome (1

  r ≤ q _{s-1 s .}

  ≤ s ≤ n) such that q < r ≤ q E. Roulette wheel selection

  Roulette wheel selection was selecting a new population with respect to the probability distribution based on their fitness values. The roulette wheel selection can be constructed as follows: The roulette wheel selection can be constructed as follows: Calculate the fitness value of each chromosome s:

  f _s

  (18)

  f  f (x ) _s

  Calculate the total fitness of population: _{pop_size}

  F  f _s  _{s k  1} (19)

  Calculate the selection probability p of each chromosome:

  f _s p  _s

  (20) _s F Calculate the cumulative probability q of each chromosome s: _s

  

  q p _{s i} 

  (21) _i  ₁

  n = Number of population s = 1, 2, 3… n F. Mutation

  There are two important things in mutation i.e determine which chromosome for mutation and mutation processes. To determine a chromosome in-mutation is started by generate random number as many as the number of population and multiple number of P-arm. Random number is generated and compared with the value of mutation rate. If the random number is less than the mutation rate, then the chromosome is selected for mutation process. The process of mutation is to exchange portions of a chromosome in the same chromosome are eligible for in-mutation. There are two mutation methods used i.e. pairs exchange and shift neighbourhood mutation. Both of the mutation applied for comparing the best fitness in the end of generation. Fig.1 and Fig.2 are the description of the two mutations used for this research. Panessai, Sanusi, Wahyudiono, Pawawo 71

  Fig.1. Pairs exchange mutation Fig.2. Shift neighbourhood mutation

  Fig.3. Multi cut point crossover

G. Crossover

  The type of crossover method used is multi cut point crossover. Fig.3 is the description of the multi cut point crossover.

  Comparison of GA-Shift Neighbourhood Mutation and GA-Pairs Exchange 72 Mutation with Multi Cut Point Crossover in Solving Rich-VRP

  3. EXPERIMENT DESIGN In order to show the effectiveness of GA, simulations were carried out. The algorithm proposes coded in java and, using a Intel(R) Core(TM) i5 CPU M430 @ 2.27GHz. As methods compared have a stochastic behaviour, they have been tested 50 times on each benchmark for every GA operator.

  TABLE

  1. BENCHMARK

  

Number of

Benchmark Benchmark Port Vehicle Vehicle Customer Fuel

  40c-9d-8k

  40

  9 8 1,275,883 28c-9d-9k

  28

  9 9 2,375,323 45c-11d-11k

  45

  11 11 3,868,567 32c-4d-8k

  32

  4 8 1,036,758 34c-11d-11k

  34

  11 11 2,743,105 63c-14d-11k

  63

  14 11 4,755,085 18c-6d-8k

  18

  6 8 1,491,149 28c-6d-11k

  28

  6 11 2,134,324 12c-4d-8k

  12

  4 8 1,263,833 53c-12d-11k

  53

  12 11 2,945,322 24c-5d-10k

  24

  5 10 1,267,387 Determination of combinations of parameter values; the probability of crossover, the probability of mutation, and the number of populations, is generally determined by "random" through experimental methods throughout the study Kumar & Panneerselvam [26] and by Bae & Moon [27]. The parameters of the genetic algorithm at one particular value were also determined by Kocha, et.al. [28]. The probability of crossover is set to 0.7; 0.8; 0.9 and 0.95 (high), and the mutation probability value set at 0.05; 0.2 and 0.3 (low), as well as with a population of 50 & 100.

  The mutation probability values to vary by 0; 0.1; 0.2 so that 1 (from the lowest to the highest), and the number of populations varies from 8 to 100. Then, the combination of the mutation probability values and the number population set one by one, while the value for crossover probability is set to 0 by Mungwattana [29] and also proposed by Volna [30]. While Ghani, et.al. [31] use very small mutation probability values (ie below 0.1), but with a large probability of crossover ranges from 0.9).

  In this research, GA parameters used are population size: 100, maximum generation: 1000, crossover rate: 0.7, mutation rate: 0.5 and selection rate: 0.5. In addition, roulette wheel selection, multi cut point crossover and two type of mutation used, namely pairs exchange mutation and shift neighbourhood mutation.

4. RESULTS AND DISCUSSION

  To check the quality of solution obtained over algorithms in 11 benchmarks then check efficiency was done. In this section, a computational study is carried out to study about performance of GA (GA Var.3 and GA Var.4) compared to the best know result and heuristic for solving our problem. Panessai, Sanusi, Wahyudiono, Pawawo 73

  TABLE

  2. FUEL CONSUMPTION OVER

  11 BENCHMARKS

  C od e Benchmark Fuel Consumption Best Known (real life) Heuristic Genetic Algorithm GA Var.3 GA Var.4

  a 40c-9d-8k 1,275,883 1,122,712 1,057,194 1,067,350 b 28c-9d-9k 2,375,323 2,064,836 1,882,530 1,900,057 c 45c-11d-11k 3,868,567 3,340,013 3,002,101 3,029,385 d 32c-4d-8k 1,036,758 919,118 879,817 888,494 e 34c-11d-11k 2,743,105 2,377,556 2,157,032 2,177,226 f 63c-14d-11k 4,755,085 4,095,004 3,665,689 3,699,593 h 18c-6d-8k 1,491,149 1,308,901 1,219,959 1,231,527 j 28c-6d-11k 2,134,324 1,858,045 1,700,970 1,716,738 k 12c-4d-8k 1,263,833 1,114,330 1,049,979 1,060,108 l 53c-12d-11k 2,945,322 2,549,070 2,307,337 2,328,831 m 24c-5d-10k 1,267,387 1,116,445 1,051,746 1,061,909

  GA Var.3 = Shift Neighbourhood Mutation GA Var.4 = Pairs Exchange Mutation The quality of solution obtained by GA in 11 benchmarks was checked by:

  100%

  X | | Best know solution Alg. propo ution known sol sed - Best

  Efficiency Algorithm 

  (22) Based on the Table 3, the average of efficiency algorithm over 11 benchmarks between roulette wheel selection, multi cut point crossover, and shift neighbourhood mutation (GA Var.3) is about 19.4350%. and roulette wheel selection, multi cut point crossover, and pairs exchange mutation (GA Var.4) is about 18.6738%. It seems the best performance of GA algorithm by roulette wheel selection, multi cut point crossover, and shift neighbourhood mutation (GA Var.3).

  TABLE.3. PERCENTAGE OF EFFICIENCY FOR FUEL CONSUMPTION OVER

  11 BENCHMARKS

  Code Benchmark Fuel Consumption Best Known (real life) Heuristic Genetic Algorithm GA Var.3 GA Var.4

  a 40c-9d-8k 12.0051 17.1402 16.3442 b 28c-9d-9k 13.0714 20.7464 20.0085 c 45c-11d-11k 13.6628 22.3976 21.6923 d 32c-4d-8k 11.3469 15.1377 14.3007 e 34c-11d-11k 13.3261 21.3653 20.6291 f 63c-14d-11k 13.8816 22.9101 22.1971 h 18c-6d-8k 12.2220 18.1867 17.4109 j 28c-6d-11k 12.9446 20.3041 19.5652 k 12c-4d-8k 11.8293 16.9211 16.1196 l 53c-12d-11k 13.4536 21.6609 20.9312 m 24c-5d-10k 11.9097 21.6609 16.2127

  Average 12.6957 19.4350 18.6738 GA Var.3 = Shift Neighbourhood Mutation GA Var.4 = Pairs Exchange Mutation B. Nag, B. L. Golden, and A. Assad, “Vehicle Routing with Site Dependencies,” Vehicle routing: methods and studies, pp.149 –159, 1988. [5]

  C. D. Tarantilis, C. T. Kiranoudis, and V. S. Vassiliadis, “A List-Based Threshold

  Heterogeneous Fleet Vehicle Routing Problem,” European Journal of Operational

  Application to a Large- Scale Real Case,” Journal of Mathematical Modelling and

  C. Prins, “Efficient Heuristics for the Heterogeneous Fleet Multi Trip VRP with

  the Operational Research Society, vol.152, pp.148 –158, 2004 [7]

  Metaheuristic for The Heterogeneous Fixed Fleet Vehicle Routing Problem,” Journal of

  C. D. Tarantilis, C. T. Kiranoudis and V. S. Vassiliadis, “A Threshold Accepting

  [6]

  Accepting Metaheuristic for the Heterogeneous Fixed Fleet Vehicle Routing Problem,” Journal of the Operational Research Society, vol.54, pp.65 –71, 2003.

  Comparison of GA-Shift Neighbourhood Mutation and GA-Pairs Exchange Mutation with Multi Cut Point Crossover in Solving Rich-VRP 74

  Research, vol.221, pp.285 –295, 2012 [4]

  A. Subramanian, P. H. V. Penna, E. Uchoa and L. S. Ochi, “A hybrid algorithm for the

  C. Prins, “Two Memetic Algorithms for Heterogeneous Fleet Vehicle Routing Problems,” Eng.Appl.Artif.Intell. vol.22, pp.916

  2009. [3]

  A. Pessoa, E. Uchoa and M. Poggi de Aragão, “A Robust Branch-Cut-and-Price Algorithm for the Heterogeneous Fleet Vehicle Routing Problem,” Network, vol.54, pp.167–177.

  International Conference on Industrial Engineering and Operations Management, pp.948- 957, 2016. [2]

  Heterogeneous Fleet VRP with Various Constraints”. Proceedings of the 2016

  A. Mungwattana, K. Soonpracha and T. Manisri, ”A Practical Case Study of a

  REFERENCES [1]

  ACKNOWLEDGMENT This work was supported in part by Grant RG078-11ICT.

  Result shows the best performance algorithm is GA Var.3. Route generated by GA Var.3 could decrease fuel consumption about 19.4350% compared to GA Var.2 about 18.6738%. This phenomenon proves that the GA proposed effectively used for solve our problem. Which the effective operator used are roulette wheel selection, multi cut point crossover, and shift neighbourhood mutation.

  5. CONCLUSIONS In this paper, the MDVRP, HFMVRP, SDCVRP and AVRP were studied and it is combined to solve ship routing. The best routing is minimum fuel consumption, maximum number port of call and maximum load factor. In order to validate our algorithms, the mathematical programming model was applied to 11 benchmarks. A computational study is carried out to compare of fuel consumption between roulette wheel selection, multi cut point crossover, and shift neighbourhood mutation to roulette wheel selection, multi cut point crossover, and pairs exchange mutation.

  Algorithms, vol.1, pp. 135 –150, 2002. [8]

• –928, 2009 [9]

  E. Choi and D. W. Tcha, “A Column Generation Approach to the Heterogeneous Fleet

  Vehicle Routing Problem,” Computers and Operations Research, vol. 34, pp.2080–2095, 2007.

  [10]

E. D. Taillard, “A Heuristic Column Generation Method for the Heterogeneous Fleet VRP,” RAIRO, vol.33, pp. 1–14, 1996.

• – 120002.

  the Heterogeneous Fleet Vehicle Rout ing Problem,” Journal of Heuristics, 2011.

  Problem, ”Computers and Operations Research, vol.38, pp.140–151, 2011

  [22] J. F. Cordeau and G. Laporte, “A Tabu Search Algorithm for the Site Dependent Vehicle

  Routing Prob lem with Time Windows,” INFOR, vol. 39, pp.292-300, 2001

  [23] L. S. Ochi, D. S. Viana, L. Drummond, and A. O. Victor, “A Parallel Evolutionary

  Algorithm for the Vehicle Routing Problem with Heterogeneous Fleet,” Future Generation Computation System, vol.14, pp.285 –292, 1998.

  [24] M. Gendreau, G. Laporte, C. Musaraganyi and E. D. Taillard, “A Tabu Search Heuristic

  for the Heterogeneous Fleet Vehicle Routing Problem,” Computers and Operations Research, vol.26, pp.1153 –1173, 1999.

  [25] N. E. A. Ghani, S. R. Shariff and S. M. Zahari, ”An alternative algorithm for VRP with time windows for daily deliveries,” Advances in Pure Mathematics,pp.342-350, 2016.

  [26] P. H. V. Penna, A. Subramanian, and L. S. Ochi, “An Iterated Local Search Heuristic for

  [27] R. Dondo and J. Cerdá, “A Cluster-Based Optimization Approach for the Multi-Depot

  I. Yusuf, M. S. Baba and N. Iksan, “Applied Genetic Algorithm for Solving Rich VRP”, Applied Artificial Intelligence, Vol. 28, No.10, pp.957-991, 2014. [21]

  Heterogeneous Fleet Vehicle Routing Problem With Time Window,” European Journal of Operational Research, vol.176, pp.1478-1507, 2007.

  [28] S. Basu, R. S. Gajulapalli, and D. Ghosh, “Implementing Tabu Search to Exploit Sparsity in ATSP Instances,” Indian Institute of Management, W.P,2008.

  [29] S. N. Kumar and R. Panneerselvam, ”Development of an efficient Genetic Algorithm for

  the time depende nt VRP with time windows”. American Journal of Operations Research.

  vol.7. pp.1-25, 2017. [30]

  X. Li, P. Tian, and Y. P. Aneja, “An Adaptive Memory Programming Metaheuristic For

  the Heterogeneous Fixed Fleet Vehicle Routing Problem,” Transport. Res.Pt.E, vol.46, pp.1111 –1127, 2010.

  [31]

  Y. Nagata and O. Bräysy,

  J. Brandão, “A Tabu Search Algorithm for The Heterogeneous Fixed Fleet Vehicle Routing

  [20]

  Panessai, Sanusi, Wahyudiono, Pawawo 75

  [15]

  [11]

  E. Volna, ”Genetic Algorithm for the VRP,” Proceeding of International Conference of Numerical Analysis and Applied Mathematics ICNAAM 2015, vol. 4. pp.120002.1

  [12]

  F. Li, B. L. Golden, and E. Wasil, “A Record-to-Record Travel Algorithm for Solving the

  Heterogeneous Fleet Vehicle Routing Problems,” Computers and Operations Research,

  vol.34, pp.2734 –2742, 2007 [13]

  G. B. Dantzig and J. H. Ramser, “The Truck Dispatching Problem,” Management Science, vol. 6, pp.80-91, 1959. [14]

  H. Bae and I. Moon, ”Multi-depot VRP with time windows considering delivery and

  

installation vehicles”. Applied Mathematical Modelling, vol.40. pp. 6536–6549, 2016

  H. C. W. Lau, T. M. Chan, W. T. Tsui, and W. K. Pang, “Application of Genetic Algorithms

  Site-

Dependent Vehicle Routing Problem,” INFOR, vol.37, pp.319-355, 1999.

  to Solve the Multi Depot Vehicle Routing Problem,” Transactions on Automation Science and Engineering, vol. 7, pp. 383-392. 2010.

  [16]

  H. Kocha, T. Henke, and G. Wäscher, ”A Genetic Algorithm for the multi-compartment

  VRP with flexible compartment sizes.” Working paper no. 4. Faculty of economics and

  management, universität magdeburg, 2016. [17]

  H. Yaman, “Formulations and Valid Inequalities for the Heterogeneous Vehicle Routing Problem,” Mathematical Programming, vol. 106, pp. 365–390, 2006. [18]

  I. C. Choi, S. I. Kim and H. S. Kim, “A Genetic Algorithm with a Mixed Region Search for

  the Asymmetric Travelling Salesman Problem,” Computers and Operations Research,

  vol.30, pp. 773 –786, 2003 [19]

  I. M. Chao, B. Golden , and E. Wasil, “A Computational Study of a New Heuristic for the

  “Edge Assembly-Based Memetic Algorithm for the Capacitated Vehicle Routing Problem,” Networks, vol.54, pp. 205–215, 2009 .